基于均值偏移粒子滤波的自适应跟踪

针对视频中运动人体的跟踪,提出了一种基于均值偏移粒子滤波的自适应跟踪算法。该算法首先对所要跟踪的人体目标进行分块,并选择与周围环境颜色相似度最小的块模板作为跟踪区域;然后使用基于均值偏移的粒子滤波方法进行跟踪,并设计了自适应更新块模板尺度的方法;最后在粒子滤波的状态估计阶段后,加入自适应观测模型,根据块模板尺度的变化情况,自适应地选择高斯噪声方差和粒子数目。实验证明,在出现遮挡或人体运动方向改变的情况下,本文算法的跟踪效果比传统均值偏移粒子滤波更好。

Adaptive tracking algorithm based on particle filter-mean shift

An adaptive tracking algorithm based on the combination of particle filter and mean shift is proposed.First,the tracking target is segmented into different block templates,and the block template with the least color similarity to the surroundings is chosen as a tracking region.Then,the block template is adjusted adaptively to the best scale when it is tracked by the proposed particle filter method based on the mean shift.Finally,an adaptive observation model is introduced after the state estimation of particle filter,and the Gaussian noise variance and the number of particles are determined according to the change of the block template scale.Experimental results show that the performance of the proposed algorithm is better than traditional method when target body is partially occluded or its moving direction is changed.